Originally Posted by Brutuz
It's a niche Intel does well in, not really much of a reason to get an Intel CPU, I'd wager more people compress files or transcode video than fold, I'm curious to see if they've tried more recent comparisons for other compilers because 2x sounds a bit much, but if it's Fortran that could be why.
But like I said before, if we talk personal uses we'll be arguing all day.
I was somewhat joking, but if you're folding, Intel CPU's are going to give you the same or higher performance usually, with much lower power consumption. For folders, this is a HUGE factor to consider, since their CPU's are at full load 24/7.
His complaint was that F@H uses the Intel compiler, and my response was ignored. Notice the AMD compiler is licensed for AMD CPU's only?
and Intel's compiler is twice as fast and more optimized, so it's no wonder they went with the Intel compiler over AMD's.
The solution here is simple, AMD needs to provide a faster and more optimized compiler over Intel's, and it also needs to support Intel CPU's. Will we see that? If the answer is no, then people need to stop complaining about companies using Intel's compiler, no matter how unfair it may be.
Besides, in the court settlement, wasn't Intel supposed to remove that limitation? Rumor has it, they haven't, and if that's the case, that would be a breach of the settlement and AMD should have a case, right? I'm not sure what the legalities of that are.
Originally Posted by Homeles
I could definitely see 30% or higher being possible, but I just don't think that AMD is much of a miracle maker, and I have to build a certain amount of humbling into my predictions to account for that. Decode heavy applications should benefit very nicely, though. Not sure what those applications would be, although I know that they're traditionally not associated with consumer workloads.
Disclaimer: I am totally pulling these numbers out of my ass, but they sound good to me
Is that bandwidth going to get utilized though? I can't see it being an issue, even for IGPs.
Originally Posted by Brutuz
I think 20% minimum, even more for multi-threaded applications too.
As for the extra bandwidth, it will increase performance on the FM* pretty decently in the real world as PC OEMs tend to bundle them with 1333Mhz RAM, and the CPU and GPU communicate through the IMC. At the very least, you can bet AMDs lead in compression will increase.
I'm gonna go out on a limb and say that Vishera is what the original Bulldozer should've been.
One of my biggest problems leading up to the initial Bulldozer launch was that I questioned whether AMD could deliver the numbers being thrown around. I'm talking about the up to 50% higher performance thanks to the 33% extra cores
I questioned if this was indeed true ( and got flamed for it often
Based on what we know of the fixes and improvements coming with SR, ( Homeles summed it up nicely ) I agree that 20% minimum
is definitely what we can expect with SR. Each core getting its own decoder alone will make a HUGE difference. It's also best to have a conservative estimate and not set our expectations too high.
Pretty sure Homeles posted this already, but just in case it's a great read:
It's laughable that AT is considered Intel biased. Some of the most detailed articles, on current and future architectures from both sides are from Anandtech. They always report and tell it how it is, something I love to do, even if people don't want to hear it. When looking at their test bench, just ignore Sysmark results, the fact that they include Sysmark does not make them biased, it's just something they've always done.
No matter if you're an Intel or AMD "fan", I think we all have a mutual goal and want a much more "All Around" competitive AMD, this benefits all of us.